KR20140090648A - Method and device for providing compensation offsets for a set of reconstructed samples of an image - Google Patents

Abstract

Compensation offsets for the reconstructed sample set of images are provided. Each sample has a sample value. A method of providing compensation offsets includes selecting a classification from a plurality of predetermined classifications based on a rate distortion criterion. Each of the predetermined classifications having a classification range that is smaller than the entire range of sample values and is made up of a plurality of classes, each class defining a range of sample values within the classification range, If it is within the scope of this related class, the sample is placed in that class. A compensation offset for applying to the sample value of each sample of the class is associated with each class of the selected classification.

Description

FIELD OF THE INVENTION [0001] The present invention relates to a method and apparatus for providing compensation offsets for a reconstructed set of samples of an image,

The present invention relates to a method and apparatus for providing compensation offsets for a reconstructed sample set of images. The present invention also relates to a method and apparatus for encoding or decoding digital video sequences.

The present invention can be applied in the field of digital signal processing and in the field of video compression using motion compensation to reduce spatial and temporal redundancy, particularly in video streams.

Many video compression formats such as H.263, H.264, MPEG-1, MPEG-2, MPEG-4 and SVC, for example, use block-based discrete cosine transform (DCT) Compensation is used. These are often referred to as predictive video formats. Each frame or image of the video signal is divided into slices that can be independently encoded and decoded. A slice is typically a rectangular portion of a frame, or more generally, a frame or a portion of an entire frame. In addition, each slice can be divided into macroblocks (MB), and each macroblock is further divided into blocks (typically blocks of 64x64, 32x32, 16x16 or 8x8 pixels).

In HEVC (High Efficiency Video Coding), 64x64 to 4x4 blocks can be used. This partitioning is organized according to a quad-tree structure based on the LCU (largest coding unit). The LCU corresponds to a 64x64 square block. If the LCU needs to be divided, the split flag indicates that the LCU is divided into four 32x32 blocks. In the same way, if any of these four blocks needs to be segmented, the segmentation flag is set to true, the 32x32 block is divided into four 16x16 blocks, and so on. When the segmentation flag is set to false, the current block is a coding unit (CU). The CU has a size of 64x64, 32x32, 16x16 or 8x8 pixels.

(Referred to as INTRA prediction) based on two series of coding modes for coding blocks of an image, a coding mode based on temporal prediction (INTER, Merge, Skip (Skip)]. In both the spatial prediction mode and the temporal prediction mode, the residual is calculated by subtracting the prediction from the original block.

An INTRA block is generally predicted from an encoded pixel at its causal boundary by an intra prediction process. In intra prediction, the prediction direction is encoded.

The temporal prediction is to find an image portion or reference area closest to the block to be encoded in a reference frame that is a previous or a future frame of the video sequence . This step is commonly referred to as motion estimation. Next, a block to be encoded is predicted using a reference region in a step, usually referred to as motion compensation. The difference between the block to be encoded and the reference portion is referred to as a motion vector indicating a reference region to be used for motion compensation Is encoded along with the item of motion information for -. In temporal prediction, at least one motion vector is encoded.

In order to further reduce the cost of encoding the motion information, rather than directly encoding the motion vector, it is assumed that the motion is homogeneous, and the motion that is calculated from one or more motion vectors of the blocks typically surrounding the block to be encoded A difference motion vector between the vector and the motion vector predictor can be encoded.

In H.264, for example, motion vectors are generated from motion vectors located in the causal neighborhood of the block to be encoded, e.g., from three blocks located above and to the left of the block to be encoded , And a median predictor to be computed. To reduce the encoding cost, only the difference between the intermediate predictor and the current block motion vector (referred to as residual motion vector) is encoded into the bitstream.

Encoding using the residual motion vector saves some bit rate, but requires the decoder to perform the same calculation of the motion vector predictor in order to decode the value of the motion vector of the block to be decoded.

Both the encoding process and the decoding process may include a decoding process of the encoded image. This process is typically performed on the encoder side for future motion estimation, which allows the encoder and the corresponding decoder to have the same reference frame.

To reconstruct the coded frame, the residual is inversely quantized and inversely transformed to provide a "decoded" residual in the pixel region. The first reconstruction is then filtered by one or several kinds of post filtering processes. These post filters are applied to the reconstructed frame on the encoder side and on the decoder side so that the same reference frame is used on both the encoder side and the decoder side. The goal of this post filtering is to eliminate compression artifacts. For example, H.264 / AVC uses a deblocking filter. This filter can remove blocking artifacts due to DCT quantization and block motion compensation of residuals. In the current HEVC standard, three types of loop filter-deblocking filter, sample adaptive offset (SAO), and adaptive loop filter (ALF) are used.

1 is a flow chart illustrating the steps of a loop filtering process of a known HEVC implementation. In an initial step 101, the encoder or decoder generates a reconstruction of the entire frame. Then, at step 102, a deblocking filter is applied to this first reconstruction to generate a deblocked reconstruction 103. [ The goal of the deblocking filter is to remove block artifacts generated by residual quantization and block motion compensation or block intra prediction. These artifacts are visually significant at low bit rates. The deblocking filter operates to smooth the block boundary according to the characteristics of two neighboring blocks. The encoding mode of each block, the quantization parameters used for residual coding, and neighboring pixel differences at the boundary are considered. The same criterion / classification is applied for all the frames, and no additional data is transmitted. The deblocking filter improves the visual quality of the current frame by removing blocking artifacts and also improves motion estimation and motion compensation for subsequent frames. In practice, the high frequencies of the block artifacts are removed, and therefore these high frequencies need not be compensated for the texture residuals of subsequent frames.

After the deblocking filter, deblocked reconstruction is filtered by a sample adaptive offset (SAO) loop filter in step 104. The resulting frame 105 is then filtered and filtered by an adaptive loop filter (ALF) at step 106 to generate a reconstructed frame 107 that is displayed and used as a reference frame for subsequent inter frames.

The goal of the SAO loop filter and ALF is to improve frame reconstruction by sending additional data, unlike deblocking filters where no information is transmitted.

The principle of the SAO loop filter is to classify each pixel as a class and add the same offset value to each pixel value of each pixel of the class. Thus, one offset is transmitted for each class. SAO loop filtering provides two kinds of classification - edge offset and band offset - for the frame region.

The edge offset classification involves determining the class for each pixel by comparing the corresponding pixel value of each pixel with the pixel value of the two neighboring pixels. Furthermore, two neighboring pixels depend on the parameter indicating the direction of two neighboring pixels. These directions are 0 degrees (horizontal direction), 45 degrees (diagonal direction), 90 degrees (vertical direction), and 135 degrees (second diagonal direction). In the following, these directions are referred to as "types" of the edge offset classification.

The second type of classification is a band offset classification that depends on pixel values. The class at the SAO band offset corresponds to a range of pixel values. Thus, the same offset is added to all pixels having pixel values within a given pixel value range.

To be more adaptive to frame content, it has been proposed to apply SAO filtering based on quad-tree structure to encode SAO. As a result, the frame region corresponding to the leaf node of the quadtree may or may not be filtered by the SAO so that only certain regions are filtered. Moreover, when SAO is enabled, only one SAO classifier - an edge offset or a band offset - is used according to the relevant parameters transmitted for each classification. Finally, for each SAO leaf node, SAO classification as well as its parameters and offsets of all classes are transmitted.

The main advantage of a quadtree is to efficiently follow the local characteristics of the signal. However, this requires dedicated encoding in the bitstream. Another solution to replace the quadtree-based encoding of the SAO parameters with the encoding at the LCU level can also be expected.

An image of the video data to be encoded can be provided as a two-dimensional array of sample values (also referred to as a color channel), each of which has a luma brightness and a neutral gray color, Represents the intensity of a color component such as a scale of chroma color deviation toward red (YUV) or a measure of intensity of red, green or blue light component intensity (RGB). The YUV model defines a color space with one luma (Y) component and two chrominance (UV) components. Generally, Y represents the luma component (luminance), and U and V are the chrominance (chroma) component.

SAO filtering is typically applied independently to the luma component and to both the U and V chroma components.

A known implementation of the SAO band offset divides the pixel value range into 32 predefined ranges of the same size as shown in Fig. The minimum value of the pixel value range is systematically 0, and the maximum value depends on the bit-depth of the pixel values according to the following relationship Max = 2 ^Bitdepth -1. For example, when the bit depth is 8 bits, the maximum value of the pixel may be 255. Thus, the pixel value range is between 0 and 255. For this 8 bit depth, each class includes a range of 16 pixel values. Moreover, for SAO band offsets, two class groups are being considered. The first group contains 16 consecutive classes in the middle of the pixel value range, as shown in gray in Fig. The second group also includes 16 classes, but these classes are at both ends of the pixel value range as indicated by the hatch in Fig. For the SAO band offset of the frame region, the group used for classification and 16 offsets are inserted into the bitstream.

3 is a flow chart illustrating steps of a method for selecting an offset in an encoder for a current frame region 303. FIG. The frame region contains N pixels. In an initial step 301, the variables Sum _j and SumNbPiX _j are set to a value of 0 for each of the 16 ranges. j represents the current range or class number. Sum _j is the sum of the differences between the values of the pixels in the range j and the values of its corresponding original pixels. SumNbPiX _j represents the number of pixels in the range j.

사이의 차가 Sum_j에 가산된다. 그 다음 단계에서, 프레임 영역(303)의 다른 픽셀들에 분류를 적용하기 위해 카운터 변수 i가 증가된다. 단계(309)에서, 프레임 영역(303)의 N개의 픽셀들 모두가 분류되었는지(즉, i≥=N인지) 여부가 판정되고, 예인 경우, 오프셋 선택 알고리즘의 최종 결과로서 각각의 클래스 j에 대한 오프셋을 제공하는 오프셋 테이블(311)을 생성하기 위해 단계(310)에서 각각의 클래스에 대한 Offset_j가 계산된다. 이 오프셋은 클래스 j의 픽셀의 픽셀 값과 그 각자의 원래의 픽셀 값 사이의 차의 평균으로서 계산된다. 클래스 j에 대한 Offset_j는 이하의 식에 의해 주어진다:In step 302, the counter variable i is set to a value of zero. Then, at step 304, the first pixel of the frame area 303 is extracted. It is assumed that the current SAO group being processed (first or second as shown in FIG. 2) is known. If it is determined in step 305 that the pixel value P _i is not in the current SAO group, the value of the counter variable i is incremented in step 308 to classify subsequent pixels of the frame area 303. Otherwise, if it is determined in step 305 that the pixel value P _i is in the current SAO group, then in step 306, a range number (or class number) j corresponding to the value of P _i is found. In a subsequent step 307, the corresponding SumNbPiX _j variable is incremented, and P _i and its original value

Is added to Sum _j . In the next step, the counter variable i is incremented to apply the classification to other pixels in the frame region 303. [ At step 309, it is determined whether all of the N pixels of the frame region 303 have been classified (i.e., whether i? = N), and if yes, as the final result of the offset selection algorithm, Offset _j for each class is calculated at step 310 to generate an offset table 311 that provides an offset. This offset is calculated as the average of the differences between the pixel values of the pixels of class j and their respective original pixel values. Offset _j for class j is given by the following equation:

4 is a flow chart illustrating the steps of a decoding process for applying a SAO band offset to a corresponding class group. This process can also be applied at the encoder side to generate reference frames that are used for motion estimation and compensation of subsequent frames.

The initial step 401 of the process entails decoding the offset values for the pixel values of each class to produce the offset table 402. [ On the encoder side, the offset table 402 is the result of the selection algorithm shown in FIG. As such, on the encoder side, step 401 is replaced by the offset selection algorithm of Fig.

In step 403, the counter variable i is set to zero. At step 405, a pixel P _i is extracted from a frame area 404 containing N pixels. In step 406, it is determined if pixel P _i belongs to the current class group. If the pixel P _i is determined to belong to the current class group, the associated class number j is identified and at step 409 the associated offset value Offset _j is extracted from the offset table 402. In step 410, the extracted offset value Offset _j is then added to the pixel value of P _i to generate the filtered pixel value P ' _i in step 411. At step 412, at the corresponding pixel, the filtered pixel value is then inserted into the filtered frame area 415. [

In step 406, if pixel P _i is not in the SAO band offset group, then in step 412 the pixel value of P _i is filtered into the filtered frame area 415. After step 412, the counter variable i is incremented, if necessary, to filter subsequent pixels of the current frame area 404. After the determination at step 414 that all N pixels of the frame region have been processed (i. E., = N), the filtered frame region 415 is reconstructed and the SAO reconstructed frame (See frame 105).

A disadvantage of the known compensation selection process is that it does not adapt to different image pixel content variations and to different types of image pixel components.

The present invention has been made in order to solve one or more of the above problems.

According to a first aspect of the present invention there is provided a method of providing compensation offsets for a reconstructed set of samples of an image, each sample having a sample value,

Selecting a classification from among a plurality of predetermined classifications based on a rate distortion criterion, each of the predetermined classifications having a classification range that is less than the entire range of sample values, Wherein each class defines a range of sample values within the classification range and wherein a sample is placed in the class if the sample value of the sample is within the range of the associated class; And

And associating a compensation offset for applying to the sample value of each sample of the associated class with each class of the selected classification.

According to a second aspect of the present invention, there is provided a method of encoding an image consisting of a plurality of samples,

Encoding the samples;

Decoding the encoded samples to provide reconstructed samples;

Performing loop filtering on reconstructed samples, wherein the loop filtering comprises applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Are provided according to a method for implementing the first aspect of the present invention; And

And generating a bit stream of encoded samples.

According to a third aspect of the present invention, there is provided a method of encoding an image comprising a plurality of samples, each sample having a sample value,

Receiving the encoded sample values;

Receiving encoded classification data;

Receiving encoded compensated offsets;

Decoding the classification data and selecting a classification from among a plurality of predetermined classifications based on the decoded classification data, each of the predetermined classifications having a classification range that is smaller than the entire range of sample values, Wherein each class defines a range of sample values within the classification range and a sample is placed in the class if the sample value of the sample is within the range of the associated class;

Decoding the encoded samples and decoding the encoded compensation offsets to provide reconstructed sample values;

Associating the decoded compensation offsets with the classes of the selected class, respectively; And

Performing loop filtering on the reconstructed sample values, wherein the loop filtering includes applying a decoded compensation offset associated with each class of the selected class to the reconstructed sample values in the range of the class of interest .

According to a fourth aspect of the present invention there is provided a signal carrying an information dataset for an image represented by a video bitstream, the image comprising a set of reconfigurable samples, Sample value, wherein the information data set includes classification data relating to a classification selected by the encoder from among a plurality of predetermined classifications, each of the predetermined classifications having a classification range smaller than the entire range of sample values, Each class defining a range of sample values within the classification range, and if the sample value of the sample is within the range of the associated class, the sample is placed in the class, and each of the plurality of classes of the selected classification The class is a boolean to apply to the sample values of the reconfigurable samples within the scope of the associated class. It is associated with an offset.

According to a fifth aspect of the present invention there is provided an apparatus for providing compensation offsets for a reconstructed sample set of images, each sample having a sample value,

Means for selecting a classification from among a plurality of predetermined classifications based on a rate distortion criterion, each said predetermined class having a classification range smaller than the entire range of sample values and comprising a plurality of classes, Define a range of sample values within the classification range, and if the sample value is within the range of an associated class, a sample is placed in the class; And

And means for associating a compensation offset for applying to the sample value of each sample of the associated class with each class of the selected classification.

According to a sixth aspect of the present invention, there is provided an encoding apparatus for encoding an image composed of a plurality of samples,

An encoder for encoding samples;

A decoder for decoding the encoded samples to provide reconstructed samples;

Loop filter-loop filtering means for filtering reconstructed samples comprises offset applying means for applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Are provided by an apparatus embodying the aforementioned fifth aspect of the present invention; And

And a bitstream generator for generating a bitstream of encoded samples.

According to a seventh aspect of the present invention, there is provided an apparatus for decoding an image comprising a plurality of samples, each sample having a sample value,

Means for receiving encoded sample values;

Means for receiving encoded classification data;

Means for receiving encoded compensated offsets;

Means for decoding the classification data and selecting a classification from among a plurality of predetermined classifications based on the decoded classification data, each of the predetermined classifications having a classification range smaller than the entire range of sample values, Wherein each class defines a range of sample values within the classification range and wherein a sample is placed in the class if the sample value of the sample is within the range of the associated class;

Means for decoding encoded samples and decoding encoded compensated offsets to provide reconstructed sample values;

Means for associating decoded compensation offsets with classes of a selected class, respectively; And

Wherein the means for performing loop filtering on reconstructed sample values comprises applying a decoded compensation offset associated with each class of the selected class to the reconstructed sample values within the range of the class of interest .

An eighth aspect of the present invention provides a method of decoding an image comprising a plurality of samples, each sample having a sample value,

Receiving the encoded sample values;

Receiving encoded classification data on a classification selected by an encoder from among a plurality of predetermined classifications, each of the predetermined classifications having a classification range that is less than the entire range of sample values, Wherein each class defines a range of sample values within the classification range, and wherein a sample is placed in the class if the sample value is within the range of the associated class;

Receiving encoded compensated offsets, each associated with classes of the selected class;

Decoding the encoded sample values to provide reconstructed sample values, and decoding the encoded classification data and compensation offsets; And

Performing loop filtering on the reconstructed sample values, wherein the loop filtering includes applying a decoded compensation offset associated with each class of the selected class to the reconstructed sample values in the range of the class of interest .

The encoder may select the classification in any suitable manner, including depending on the rate distortion criterion or on the properties of the statistical distribution of the sample values.

In the context of the present invention, a sample may correspond to a single pixel, and a sample value may correspond to a respective pixel value. Alternatively, the sample may comprise a plurality of pixels, and the sample value may correspond to a pixel value determined from the pixel values of the plurality of pixels.

At least some of the methods according to the present invention may be implemented in a computer. Accordingly, the present invention may take the form of a hardware only embodiment, a software only embodiment (including firmware, resident software, microcode, etc.), or an embodiment with both software and hardware aspects, May be referred to herein as a "circuit "," module ", or "system ". In addition, the invention may take the form of a computer program product embodied in any type of tangible medium of representation in which computer usable program code is embodied.

As the present invention may be implemented in software, the present invention may be embodied in a computer readable code for provisioning in a programmable device or any suitable carrier medium. Type of transmission medium may include storage media such as floppy disks, CD-ROMs, hard disk drives, magnetic tape devices or solid state memory devices, and the like. The transient carrier medium may include signals such as electrical signals, electronic signals, optical signals, acoustic signals, magnetic signals, or electromagnetic signals (e.g., microwave or RF signals).

Thus, in accordance with a ninth aspect of the present invention, there is provided a computer program product for a programmable apparatus, the computer program product comprising program code executable on a first, second, third, And a sequence of instructions implementing a method of implementing any of the eighth aspects.

According to a tenth aspect of the present invention there is provided a computer readable storage medium storing instructions for a computer program embodying a method for implementing any of the first, second, third and eighth aspects described above / RTI >

With reference to the following drawings, and by way of example only, embodiments of the present invention will now be described.
1 is a flow chart illustrating steps of a prior art loop filtering process;
Figure 2 graphically illustrates a sample adaptive band offset classification of a prior art HEVC process;
3 is a flow chart illustrating the steps of a process for determining a compensation offset for an SAV band offset of an HEVC;
4 is a flow chart illustrating the steps of the HEVC SAO band offset filtering process;
5 is a block diagram that schematically illustrates a data communication system in which one or more embodiments of the present invention may be implemented;
6 is a block diagram illustrating components of a processing apparatus in which one or more embodiments of the present invention may be implemented;
7 is a flow chart illustrating the steps of an encoding method according to embodiments of the present invention.
8 is a flowchart illustrating steps in a loop filtering process in accordance with one or more embodiments of the present invention.
9 is a flowchart illustrating steps of a decoding method according to embodiments of the present invention.
10 is a flow chart illustrating steps of a method for determining an SAO band offset classification, in accordance with a first embodiment of the present invention.
11 is a flow chart illustrating steps of a method for determining an adapted classification, in accordance with an embodiment of the present invention.
Figure 12 is a flowchart illustrating steps of a method for determining an adapted classification, in accordance with an alternative embodiment of the present invention.
Figure 13 illustrates some of the dimensions of the validity range for classification, in accordance with an embodiment of the present invention.
Figure 14 illustrates several sizes of classes for classification, in accordance with an embodiment of the present invention.
15 is a diagram illustrating some of the sizes of classes in an effective range for classification according to an embodiment of the present invention.
Figure 16 illustrates several central positions of a first group of valid ranges for classification, according to one embodiment of the present invention.
Figure 17 illustrates several central positions of a second group of valid ranges for classification, according to an embodiment of the present invention.
Figure 18 shows a rate distortion selection of a parameter classification, in accordance with an embodiment of the present invention.
Figures 19A and 19B illustrate possible positions of an effective range within the full range, according to another embodiment of the present invention.
20A illustrates pseudo code applied in the prior art for encoding SAO parameters at the LCU level;
Figure 20B illustrates an improved pseudo code according to an embodiment of the present invention for encoding SAO parameters at the LCU level.
FIG. 21 is a flowchart corresponding to the pseudo code of FIG. 20A. FIG.
FIG. 22 is a flowchart corresponding to the pseudo code of FIG. 20B. FIG.
23 is a flowchart for use in describing the encoding of SAO parameters, in accordance with a further embodiment of the present invention.
24 illustrates pseudo code used to encode SAO parameters, in accordance with another embodiment of the present invention.
25 is a flowchart corresponding to the pseudo code shown in Fig.
26 is a flowchart for use in describing the encoding of SAO parameters, in accordance with another additional embodiment of the present invention.

5 illustrates a data communication system in which one or more embodiments of the present invention may be implemented. The data communication system includes a transmitting device (in this case, server 501) that is operative to transmit data packets of a data stream over a data communication network 500 to a receiving device (in this case, the client terminal 502) . The data communication network 500 may be a wide area network (WAN) or a wide area network (LAN). Such a network may be, for example, a wireless network (WiFi / 802.11a or b or g), an Ethernet network, an internet network, or a mixed network of several different networks. In a particular embodiment of the present invention, the data communication system may be a digital television broadcasting system in which the server 501 transmits the same data content to a plurality of clients.

The data stream 504 provided by the server 501 may comprise multimedia data representing video and audio data. In some embodiments of the invention, the audio and video data streams may be captured by the server 501 using a microphone and a camera, respectively. In some embodiments, the data stream may be stored in server 501, received by server 501, or generated at server 501 from another data provider. The server 501 has an encoder that encodes the video and audio streams to provide a compressed bitstream for transmission, in particular a more compact representation of the data presented as input to the encoder.

In order to obtain a better ratio of the quality of the transmitted data to the amount of transmitted data, the compression of the video data may, for example, follow the HEVC format or the H.264 / AVC format.

The client 502 receives the transmitted bit stream, decodes the reconstructed bit stream to reproduce the video image on the display device and the audio data by the speaker.

While the streaming scenario is contemplated in the example of FIG. 5, it should be noted that in some embodiments of the present invention, the data communication between the encoder and the decoder may be performed using a media storage device, I will know.

In one or more embodiments of the present invention, a video image is transmitted with data representing compensation offsets for applying to the reconstructed pixels of the image to provide filtered pixels to the final image.

Figure 6 schematically depicts a processing device 600 configured to implement at least one embodiment of the present invention. The processing device 600 may be a device such as a microcomputer, a workstation, or a lightweight portable device. Apparatus 600 includes a communications bus 613 coupled to:

A central processing unit 611, such as a microprocessor, indicated as a CPU;

- a read only memory (ROM) 607, which stores computer programs embodying the invention;

- a random access memory 612 indicated as RAM for storing executable code of the method of embodiments of the present invention, as well as a method for encoding a digital video sequence in accordance with embodiments of the present invention and / A register configured to record the parameters and parameters necessary to implement the method; And

A communication interface 602 (to which the digital data to be processed is transmitted or received).

Optionally, the device 600 may also include the following components:

Data storage means 604, such as a hard disk, for storing computer programs embodying methods of one or more embodiments of the invention and data used or generated during the implementation of one or more embodiments of the invention;

A disk drive 605 for the disk 606 (the disk drive is configured to read data from or write data to the disk 606);

Screen 609 which serves as a graphical interface with the user by displaying data and / or by keyboard 610 or any other pointing means.

The device 600 may be coupled to various peripherals such as, for example, a digital camera 620 or a microphone 608, each of which may include an input / output card (not shown) to provide multimedia data to the device 600, Respectively.

The communication bus provides communication and interoperability between the various components contained in or connected to the device 600. The representation of the bus is not intended to be limiting, and in particular, the central processing unit is responsible for transferring instructions to any element of the device 600, either directly or by other elements of the device 600. [

The disc 606 may be embodied as any information carrier, such as, for example, a compact disc (CD-ROM) (rewritable or otherwise), a ZIP disk or a memory card, and in general terms, A method for encoding a digital video sequence according to the present invention which may be read by a microprocessor, integrated in a device, or maybe not, and which is mobile and one or more programs - its implementation to be implemented, and / Which can be replaced by information storage means configured to store the information.

Executable code may be stored in read only memory 607, on hard disk 604, or on removable digital media (e.g., disk 606 as described above). According to one variant, the executable code of the programs is transferred to the communication network 603 via the interface 602, to be stored in one of the storage means of the device 600, such as the hard disk 604, Lt; / RTI >

The central processing unit 611 is configured to control and direct the execution of instructions stored in one of the above-mentioned storage means or portions of software code of the program or programs according to the present invention. Programs or programs stored in the non-volatile memory (e.g., on the hard disk 604 or in the read-only memory 607) upon power up are transferred into the random access memory 612 , Random access memory 612 includes executable code of a program or programs), as well as to registers that store the variables and parameters necessary to implement the invention.

In this embodiment, the device is a programmable device that uses software to implement the invention. However, as a further alternative, the present invention may be implemented in hardware (e.g., in the form of an Application Specific Integrated Circuit (ASIC)).

Figure 7 shows a block diagram of an encoder in accordance with at least one embodiment of the present invention. An encoder is shown as coupled modules and each module is operable in the form of programming instructions to be executed by, for example, the CPU 611 of the device 600, in accordance with one or more embodiments of the present invention, And at least one corresponding method of implementing at least one embodiment for encoding an image of a video stream.

The original digital video sequence io to in (701) is received as an input by the encoder 70. Each digital image is represented by a set of samples called pixels.

After the implementation of the encoding process, the bitstream 710 is output by the encoder 70. The bitstream 710 includes a plurality of encoding units or slices, each slice including a slice header for transmitting the encoding values of the encoding parameters used to encode the slice, and encoded video data And a slice body to which the slice body is attached.

The input digital images i0 to in (701) are divided into pixel blocks by a module 702. [ The blocks may correspond to image portions and may have variable sizes (e.g., 4x4, 8x8, 16x16, 32x32, 64x64 pixels). The coding mode is selected for each input block. A coding mode based on two series of coding mode-spatial prediction coding (intra prediction), and a coding mode based on temporal prediction (inter coding, merging, skipping). Possible coding modes are tested.

Module 703 implements an intra prediction process where the given block to be encoded is predicted by a predictor calculated from neighboring pixels of the block to be encoded. To provide the residual when intra coding is selected, the representation of the selected intra predictor and the difference between the given block and its predictor are encoded.

The temporal prediction is implemented by the motion estimation module 704 and the motion compensation module 705. First, a reference image is selected from the reference image set 716 and a portion of the reference image (also referred to as a reference region or an image portion), which is the region closest to the given block to be encoded, is selected by the motion estimation module 704. The motion compensation module 705 then predicts the block to be encoded using the selected region. A difference (also referred to as a residual block) between the selected reference area and a given block is calculated by the motion compensation module 705. The selected reference area is indicated by a motion vector.

Thus, in both of these cases (spatial prediction and temporal prediction), the residual is calculated by subtracting the prediction from the original block.

In intra prediction, which is implemented by module 703, the prediction direction is encoded. In temporal prediction, at least one motion vector is encoded.

When inter prediction is selected, the information on the motion vector and the residual block are encoded. To further reduce the bit rate, the motion vector is encoded by the difference for the motion vector predictor, assuming that the motion is homogeneous. The motion vector predictors of the motion information predictor set are obtained from the motion vector field 718 by the motion vector prediction and coding module 717. [

The encoder 70 further includes a selection module 706 for selecting a coding mode by applying an encoding cost criterion such as a rate-distortion criterion. To further reduce redundancy, a transform (such as DCT) is applied to the residual block by the transform module 707, the transformed data to be obtained is then quantized by the quantization module 708 and the entropy encoding module 709 Lt; / RTI > Finally, the encoded residual block of the current block being encoded is inserted into the bitstream 710.

The encoder 70 also performs decoding of the encoded image to generate a reference image for motion estimation of the subsequent image. This allows the encoder and decoder receiving the bitstream to have the same reference frame. The inverse quantization module 711 performs inverse quantization of the quantized data, and then the inverse transformation by the inverse transformation module 712 is performed. The reverse intra prediction module 713 uses the prediction information to determine which predictor to use for a given block and the reverse motion compensation module 714 actually uses the prediction information to the module 712 To the reference region obtained from the reference image set 716. [

Post filtering is then applied by module 715 to filter the reconstructed pixel frame. In embodiments of the present invention, a SAO loop filter is used in which the compensation offset is added to the pixel value of the reconstructed pixel of the reconstructed image.

8 is a flow chart illustrating the steps of a loop filtering process in accordance with at least one embodiment of the present invention. In an initial step 801, the encoder generates a reconstruction of the entire frame. Then, at step 802, a deblocking filter is applied to this first reconstruction to generate 803 a deblocked reconstruction. The goal of the deblocking filter is to remove block artifacts generated by residual quantization and block motion compensation or block intra prediction. These artifacts are visually significant at low bit rates. The deblocking filter operates to smooth the block boundary according to the characteristics of two neighboring blocks. The encoding mode of each block, the quantization parameters used for residual coding, and neighboring pixel differences at the boundary are considered. The same criterion / classification is applied for all the frames, and no additional data is transmitted. The deblocking filter improves the visual quality of the current frame by removing blocking artifacts and also improves motion estimation and motion compensation for subsequent frames. In practice, the high frequencies of the block artifacts are removed, and therefore these high frequencies need not be compensated for the texture residuals of subsequent frames.

After deblocking filter, deblocked reconstruction is filtered by a sample adaptive offset (SAO) loop filter in step 804, based on the classification 814 of pixels determined according to embodiments of the present invention. The resulting frame 805 may then be displayed and filtered by an adaptive loop filter (ALF) at step 806 to generate a reconstructed frame 807 to be used as a reference frame for subsequent interframes.

In step 804, each pixel of the frame region is classified into a class of classification determined according to its pixel value. The class corresponds to the determined range of pixel values. The same compensation offset value is added to the pixel value of all pixels having a pixel value within a given pixel value range.

The determination of the classification of pixels for sample adaptive offset filtering will be described in more detail below with reference to any of Figs. 10-17.

9 shows a block diagram of a decoder 90 that may be used to receive data from an encoder, in accordance with an embodiment of the invention. The decoder is represented by the connected modules, and each module is associated with a corresponding step of the method implemented by the decoder 90, for example in the form of programming instructions to be executed by the CPU 611 of the device 600 .

The decoder 90 receives a bitstream 901 comprising encoding units, each encoding unit comprising a body comprising a header and encoded video data including information about encoding parameters. As described in connection with FIG. 7, the encoded video data is entropy encoded, and the index of the motion vector predictor is encoded into a predetermined number of bits for a given block. The received encoded video data is entropy decoded by module 902. [ The residual data is then inverse-quantized by module 903 and the inverse transform is applied by module 904 to obtain pixel values.

Mode data representing the coding mode is also entropy decoded and, based on the mode, intra type decoding or inter type decoding is performed on the encoded video data block.

In the case of the intra mode, the intra predictor is determined by the inverse intra prediction module 905 based on the intra prediction mode specified in the bitstream.

If the mode is inter, motion prediction information is extracted from the bit stream to find the reference area used by the encoder. The motion prediction information is composed of a reference frame index and a motion vector residual. A motion vector predictor is added to the motion vector residual to obtain a motion vector by the motion vector decoding module 910. [

The motion vector decoding module 910 applies motion vector decoding to each current block encoded by motion prediction. Once the index of the motion vector predictor for the current block is obtained, the actual value of the motion vector associated with the current block may be decoded and used by module 906 to apply the inverse motion compensation. To apply the inverse motion compensation 906, a portion of the reference image represented by the decoded motion vector is extracted from the reference image 908. The motion vector field data 911 is updated with the decoded motion vector in order to be used for the backward prediction of the subsequent decoded motion vector.

Finally, the decoded block is obtained. Post filtering is applied by post filtering module 907, similar to the post filtering module 815 applied in the encoder as described with reference to Fig. The decoded video signal 909 is finally provided by the decoder 90.

10 is a flow chart illustrating steps of a method according to a first embodiment of the present invention for classifying reconstructed pixels of an image for application of a compensation offset. In this embodiment, the class for classification of the pixel according to the pixel value of the reconstructed pixel of the frame region is determined based on the statistical distribution of the reconstructed pixel values of the frame region. The center, the effective range, and the amount of pixels per class are determined based on the distribution of pixel values. In this embodiment, the decoder can apply exactly the same process as the decoder for segmentation of the distribution.

In the early stages of this process, the module 1002 scans the current frame area 1001 to determine the statistical distribution of the pixel values of the pixels of the frame area 1001 and to generate a corresponding histogram 1003. In one particular embodiment, the process includes updating a table containing the number of pixels for each pixel value (i. E., For each pixel value, the number of pixels having that pixel value is created as a table ). The table contains the number of cells equal to the maximum pixel value MAX determined according to the expression Max = 2 ^Bitdepth - 1 based on the bit depth of the pixels.

The module 1004 then determines the center of the histogram 1003 generated. The range of effective pixel values of the histogram is then determined by module 1006 based on the distribution of pixel values represented in histogram 1003 and, where appropriate, based on the center of the histogram. Finally, an equiprobable class that defines a range of pixel values is determined. As such, a table 1009 containing a range of pixel values for each class, or, alternatively, a table containing pixel values for each pixel is provided. In some embodiments of the invention, the determination of the class of the same probability may depend on the predetermined number of classes 1000.

At step 1004, various algorithms may be used to determine the center of the histogram 1003 generated. In one embodiment, the minimum value Min _Hlst and the maximum value Max _Hist of the histogram can be found. To find the minimum value Min _Hist , the cells Hist _k of the histogram are scanned from the pixel value 0 to the first cell Hist _k of the histogram, not 0. To find the Μax _Hist , the cells are scanned in reverse order (from the maximum pixel value MAX to the first cell Hist _k in the histogram, not 0). The center _Hist of the histogram is calculated as:

Center _Hist = (Max _Hist - Min _Hist ) 12 + Min _Hist

In an alternative embodiment, the center of the histogram is considered to be a weighted average center of the distribution. If the value of the histogram cell Hist _k is deemed to be the number of pixels having a value k, the Center _Hist is calculated as:

Where N is the number of pixels in the current frame region.

In step 1006, one possible technique for determining the effective range of the generated histogram is to select Min _Hlst and Max _Hist as described above for both ends of the effective range.

In another embodiment, the minimum value Min _Range of the histogram is determined by scanning from 0 to the first Hist _k having a value greater than the threshold value alpha. In the same manner, the Max _Range is determined by inverse scanning from the maximum pixel value MAX to the first Hist _k greater than the threshold value alpha. The threshold value a may be a predetermined value. As another alternative, the threshold value alpha may depend on the number of pixels in the frame region and / or on the component type (chroma and luma) of the input signal.

In one particular embodiment, the number of classes can be considered to be known on the encoder side and on the decoder side. The number of classes of pixel values may depend, for example, on the number of pixels in the current frame region according to each component (luma, chroma U and V).

To create a class with the same probability, the number of pixels NbPix _Range in the validity range 1007 is defined. The number of pixels in the effective range NbPix _Range is determined by scanning each histogram cell Hist _k from k = Min _Range to k = Max _Range . Then, to determine the optimal number of pixels in each class NbPix _classes, the predetermined number of pixels NbPix _Range in the validity range is divided by the number of classes (1000).

에 의해 식별된다.11 is a flowchart illustrating steps of an algorithm for determining a class of the same probability, according to an embodiment of the present invention. In the initial step 1101, the class number j is set to 0 and the current pixel value k is set to Min _Range . For equiprobable classification, a class is identified by its pixel value range. The class number j thus ranges from its minimum pixel value Min _j to its maximum pixel value Max _j in its range

Lt; / RTI >

가 결정된다. 2개 이상의 클래스에서 동일한 픽셀 값을 획득하는 것을 피하기 위해 단계(1109)에서 변수 k가 증가된다. 더욱이, 그 다음 클래스에 대한 픽셀 값 범위를 정의하기 위해 단계(1110)에서 변수 j가 또한 증가된다. 변수 j가 클래스 수 NbPix_classes보다 큰 경우, 단계(1112)에서 모든 클래스가 정의된 것으로 간주될 수 있다.In step 1103, the minimum pixel value Min _j of the current class indexed by _j is set to the current pixel value k. Subsequently, at step 1104, SumNbPiX _j is set to zero. SumNbPiX _j corresponds to the number of pixels in the range j. Then, at step 1105, the number of pixels (Hist _k ) having the pixel value k is added to SumNbPiX _j . In step 1106, it is determined whether the sum SumNbPiX _j of the number of pixels for the current class j is greater than the number NbPix _classes of pixels in the _classes . If this condition is not reached, the value of k is incremented in step 1107, and the number of pixels Hist _k or pixel value k is added to SumNbPiX _j in step 1105. If it is determined that SumNbPiX _j > NbPix _classes or k reaches the maximum value Max _Range of the valid range, then in step 1108 the maximum value for current class j is equal to the current value of k. In this stage, a class j is defined. That is, the range of class j

Is determined. The variable k is incremented in step 1109 to avoid obtaining the same pixel value in more than one class. Furthermore, in step 1110, the variable j is also incremented to define a range of pixel values for the next class. If the variable j is greater than the class number NbPix _classes , then in step 1112 all classes may be deemed defined.

As a result, the encoder will determine the offset value for each class j and transmit it to the decoder, as described in connection with FIG. The encoder and decoder will filter the frame area, as described with reference to Fig.

In this embodiment, it should be noted that since the number of classes is predetermined based on syntax values, the number of classes NbClasses does not depend on the pixel value. As a result, in this embodiment, the parsing of the SAO band offset is independent of the decoding of other frames. It should be noted that parsing for the SAO band offset includes parsing of each offset.

In a further embodiment of determining the classification of the same probability, the number of classes may be determined according to the distribution of pixel values in the generated histogram. In practice, when the amplitude of the effective range is high or low, the number of classes will affect the coding efficiency. As a result, a better adaptive classification can be provided by determining the number of pixels as well as the number of pixels in each class.

Figure 12 is a flowchart depicting the steps of an algorithm according to a further embodiment for providing a more adaptive classification. This flowchart is based on the flowchart of the embodiment of FIG. 11, wherein modules having the same end number perform the equivalent function. However, the decision modules 1206 and 1211 of this embodiment process test conditions that are different from the test conditions that are processed by the corresponding modules 1106 and 1111 of FIG.

In this embodiment, if SumNbPiX _j > NbPix _classes , or if k reaches the maximum value Max _Range of the validity range, or if k - Min _j is less than the MaxClass _Range for the class, then the decision module 1206 Stop the loop based on the value of k and select Max _j for class j. k - Min _j corresponds to the number of pixel values in the current range of class j. The MaxClass _Range is a predetermined maximum number of pixel values in the range. This range may depend on the bit depth, the number of pixels N in the frame region, and the type of signal (luma, chroma U and V). For example, when the bit depth is 8, the MaxClass _Range for the luma component may be 16, as in the HEVC implementation.

The advantage of the embodiment of FIG. 12 as compared to the embodiment of FIG. 11 is its coding efficiency for pixel value distributions with large amplitudes. This embodiment is more adaptive to the distribution.

It should be noted that, in this embodiment, the number of classes determined is dependent on the pixel value, and therefore the parsing of the current frame is dependent on the decoding of the previous frame. To be more robust against transmission errors, the class number NbClasses is inserted into the bitstream. The transmission of such data does not significantly affect the coding efficiency.

The main advantage of the first embodiment of the classifications of Figures 10-12 is that the classification is adapted to the pixel value distribution. Furthermore, the center, scope, size and amount of each class need not be transmitted. As a result, no additional data other than the data representing the offset of each class need to be transmitted for the determined classification, as in the known HEVC implementation.

A further embodiment of the invention for determining the classification, accompanied by signaling of the parameters of the classification, will now be described with reference to Fig. The purpose of a further embodiment of the classification is to provide an optimal classification of the distribution of pixel values. The difference compared with the previous embodiments is that the classification is determined based on the rate-distortion criterion, not determined based on the distribution of the pixel values directly. In a further embodiment, the encoder selects a classification that best fits the pixel value distribution among the predefined possible classifications. This selection is based on a rate distortion criterion. As in previous embodiments, the center, effective range, and magnitude of the classes of generated histograms representing the distribution of pixel values are determined. In a further embodiment, these parameters are transmitted in the bitstream. To minimize the impact of the transmission of such data, the size of the class and associated range is selected from predefined values. As a result, the encoder inserts the center of the selected classification, the index associated with the selected classification, and the size of the class of classification into the bitstream.

To provide adaptation to the distribution of pixel values, several pixel value range sizes are defined, as shown in FIG. In Figure 13, the entire pixel value range is divided into 32 subranges. Four examples (1301, 1302, 1303, 1304) are shown for a first class group with pixel values located in the middle of the pixel value range. The first example 1301 includes 26 ranges out of the 32 possible ranges. Thus, the effective range 1301 indicates 13/16 of the whole range. In the same way, 1302 represents only 8 of the 32 possible ranges (i.e., 1/4 of the effective range), 1303 represents 1/8 of the total range, and 1304 represents 1/16 of the entire range . For the proposed scheme, all possible sizes from the entire range to the range corresponding to only one pixel value can be considered. The number of possible valid ranges must either be predetermined according to the coding efficiency, or must be predetermined for the number of pixels in the frame area.

Figure 13 also shows some examples of sizes for a second class group with respect to pixel values located at the edge of the pixel value range. The second group includes two subclass groups-one located on each edge of the histogram. Examples 1305, 1306, 1307 and 1308 represent the same number of pixel values as the first group of examples 1301, 1302, 1303 and 1304, respectively.

In an embodiment of the present invention, the size of the class (i. E., The range of pixel values per class) is not fixed compared to the prior art method. Fig. 14 shows examples 1401 to 1406 of several sizes. In this example, the class size is from 32 pixels (1401) to only one pixel (1406). As described above with respect to FIG. 13, these class sizes may be combined with all possible valid ranges. In this embodiment, all classes are considered to have the same magnitude for a particular pixel value range. Thus, for one group, data indicating the size of the effective range and the size of the class is inserted into the bitstream.

In another embodiment, the size of the class for a given validity range is adapted according to the location of the class in the validity range. More precisely, the size of the class is adapted to the distribution of pixel values. In a further embodiment, these sizes are predetermined according to the pixel value distribution for each validity range. In practice, the histogram of the pixel value distribution generally corresponds to a Gaussian distribution. The closer the pixel value is to the center of the histogram, the more pixels have pixel values close to this value. This means that the nearest histogram cell Hist _k has a larger value (the number of corresponding pixels) than the histogram cell Hist _k at both ends of the effective range of the histogram.

Figure 15 shows an example of two described embodiments of class size. The example 1501 shows a fixed size of 8 pixel values for the valid range of 32 pixel values. And 1502 is a fixed size of four pixel values for the same effective range size.

An example 1503 illustrates another embodiment of an adaptive class size for the current range of 32 pixel values. In this example, the classes at both ends of the validity range are larger than the classes at the center (i.e., have a wider range of pixel values), and these classes have eight pixel values and two pixel values, respectively. Between these classes, two different classes have a range of four pixel values.

The size of the classes for the second group may also be adapted to the distribution of pixel values. The goal of the second group of current HEVC implementations is to use only the two ends of the histogram. In fact, both ends of the histogram contain extreme values (due to lossy coding), where the error is often related to the usually higher frequencies, as compared to the low frequencies. In the same manner as in the first group, several class sizes may be tested for the valid ranges of the second group. In that case, for the two subgroups of the second group, the subdivisions 1501 and 1502 may be compared using a rate distortion criterion.

Furthermore, an embodiment in which the size of the class is adapted may be applied. The example 1504 shows the proposed adapted class size for the first range (left side) of the second group. And the example 1505 shows the proposed adapted class size for the second subgroup (right side) of the second group. In that case, the class at both ends of the class nearer to the center contains more pixel values.

The goal of the second group is to use both ends of the histogram; As a result, it is sometimes useful to use inverse adaptation of the magnitude for the second group. In that case, an example 1504 is used for the second lower group (right side), and an example 1505 is used for the first lower group (left side) of the second group. In this embodiment, the class at both ends of the class nearer to the center contains fewer pixel values. In that case, the goal is not to generate classifications of the same probability but to find better segmentation of both ends of the second group.

Since the center of the statistical distribution of pixel values is not necessarily in the middle of the entire pixel value range, the center of the distribution based on the effective range must be determined and transmitted with the image data in the bitstream. Figure 16 shows an example of a full range with different center positions for an effective range corresponding to 1/4 of the full range. Unlike the example 1302 in Fig. 13, for the four examples 1601, 1602, 1603 and 1604 in Fig. 16, the center of the effective range is not located at the center of the entire range. This solution allows the selected classification to be adapted to the distribution of pixel values.

The determined center can then be coded for transmission in the bitstream. Several techniques for coding data may be expected.

If the bit depth of the current frame region is considered to be 8 bits, the number of positions that can be considered for the median value corresponds to 256 - the size of the minimum effective range. For example, compared to FIG. 13, the minimum size of the effective range is 2, and these two classes may include at least one pixel. Thus, for this particular example, the center may have a value from 1 to 254, thus 254 positions for the center can be considered.

Another solution is to quantify the median value. In one embodiment, the center is coded according to the size of the classes. Thus, for example, if the size of the classes (or the minimum size of all classes of the current scope when the adapted class size scheme is used) is one pixel value, then the center is not quantified and all possible centers Location. As shown in FIG. 16, when the size of classes is 16 pixel values, only pixel values of every 16 pixel values can be considered. 16, the centers for the examples 1601, 1602, 1603, and 1604 are 9, 23, 27, and 6, respectively. In other embodiments, only center positions that are a multiple of the maximum size of the classes defined in the algorithm may be considered. Thus, the center is the pixel value divided by the maximum size of the classes. This provides a reduction in terms of the number of bits to be transmitted.

Moreover, theoretically, the most probable center is the center of the entire range. Thus, the data transmitted to determine the center position at the decoder side is the difference between the center of the full range and the center of the valid range of the current classification. Thus, for example, in FIG. 16, the data transmitted in relation to the center for the examples 1601, 1602, 1603 and 1604 are 16-9 = 7, 16-23 = -7, 16-27 = -11, 16-6 = 10.

For the second group, the center of the histogram need not be coded. Thus, several schemes can be considered to code the displacement of the two subgroups for the second group. The proposed embodiments of quantizing the median value described for the first group can be easily extended to the proposed embodiments for the second group.

In embodiments of the present invention, the location of the validity range (selected classification) may be specified with the same precision or granularity over the entire range, i.e., regardless of the location of the classification within the entire range . This is the case in the examples 1601 to 1604 shown in FIG. 16, where the positions (center positions) are 9, 23, 27 and 6. The entire range is shown as 0 to 32. There are 32 possible positions, and the granularity is the same over the entire range.

However, it is also possible to provide more possible positions in other parts of the entire range in one part of the full range, as shown in Figs. 19A and 19B. In other words, the granularity of the position varies depending on where the classification is within the overall range. These embodiments provide unequal quantization of the full range (here indicated by 0 to 32) by variable granularity to more precisely locate the center of the classification (effective range) to the full range of the most important (or promising) I am proposing. In addition, unequal quantization allows the number of bits needed to signal the position of the classification to be limited, while still providing adequate precision in the critical portion of the entire range. For example, in the middle of the entire range as shown in Fig. 19A, such finer granularity can be applied. In this figure, possible central positions correspond to the indices indicated by the bold solid lines. The spacing between the two possible central positions is smaller in the middle of the entire range than at the ends. Thus, the center position in the middle of the entire range can be set more precisely than at the end of the entire range.

In Figure 19b, the spacing between the two possible central positions is smaller at both ends of the full range than in the middle. For example, this embodiment may be particularly useful when it has significant sample values at the extreme value of the distribution.

More generally, finer quantization can be applied anywhere in the entire range.

When variable quantization as described above is used, the classification range (size of the effective range) can be fixed for all positions. For example, the classification range may include four classes, each of which consists of eight pixel values.

It is also possible to vary the classification range / class size depending on the position, and thus, in Fig. 19A, the classification range is, for example, eight pixel values at positions 12 to 20, 16 pixel values at positions 10 and 26, And 28 pixel values.

The variable quantization can be used regardless of the method applied to determine the classification range, as described herein. The method can use, for example, the properties of the statistical distribution of the sample values or use a rate distortion criterion.

Variable quantization can be predetermined in the encoder as well as in the decoder. For example, the encoder and decoder may assign indices to possible center positions (or left positions), e.g., in Figure 19A, position 2 is index 0, position 6 is index 1, position 10 is index 2, position 12 is index 3, position 13 is index 4, and so on. It is then sufficient for the encoder to send the index of the selected classification to the decoder. As an alternative, information about the variable quantization may be determined at the encoder and signaled to the decoder through the bitstream.

In one particular embodiment, it can be considered that the center of the histogram is always the center of the entire range. Thus, in that case, only one displacement is considered. Both groups are scaled centrally by the same displacement. As a result, only one data-displacement of the first range of the second group-can be coded. Examples (1701, 1702, 1703 and 1704) of FIG. 17 are examples of such displacements. In the examples 1701, 1702, 1703 and 1704, the displacements are 4, 6, 10 and 0, respectively. The displacement can be directly coded without prediction.

In a further embodiment, as shown in examples 1705, 1706, 1707 and 1708, both subgroups of the second group have independent positions in the entire range. Two coding schemes can be considered.

In the first scheme, the center of the non-existent first group is coded with the size of the effective range of the first group that does not exist.

A second way of independently coding both groups is to transmit two displacements (one for each group) from the two ends of the full range. Thus, for the examples 1705, 1706, 1707 and 1708, the displacements transmitted are 11 and 32-28 = 4 for 1705, 21 and 0 for 1706, 3 and 32-16 for 1707, = 32, 7 for 1708, and 32-31 = 1.

18 is a flow chart illustrating the steps of a rate distortion selection algorithm, in accordance with an embodiment of the present invention. For the sake of simplicity, only the selection for the first group that does not have an adapted class size is considered. The selection for other embodiments described above can be readily adapted.

In the initial step 1801, the statistics of the current frame area are calculated. This involves determining the variables Hist _k and Sum _k for all pixel values k. Hist _k corresponds to the number of pixels having the same pixel value as the variable k, and Sum _k corresponds to the sum of the differences between all the pixels whose pixel value is the value k and its original pixel value. This algorithm includes three loops with respect to the three parameter-class size S, size range R of the range and center C-. In step 1803, the first loop tests each possible class size. For example, the size is defined in Fig. In step 1804, an offset for each sub-range of the entire range is calculated. For example, if the bit depth is 8 and the size of the class is 16, then the distortion and offset for the 32 possible ranges in the entire range are calculated. By properties, offsets and distortions are calculated by linear combination of Hist _k and Sum _k or all k values in the current range. Then, at step 1807, for each possible range R 1805 and for each possible center C 1806, a rate distortion cost is estimated. This evaluation is based on a rate distortion criterion. When all the central C 1808, all ranges 1809 and all sizes 1810 are tested, at step 1811, the best parameters S, R, C are selected based on the best rate distortion cost. Advantages of this second scheme of generating the same probability classifications include reduced complexity and improved coding efficiency. The classification selection of the center, range, and size of the class provides an optimal rate distortion selection, as compared to embodiments in which the classification is based on a statistical distribution of pixel values. Of course, this embodiment provides an improvement in terms of coding efficiency as compared to the current HEVC implementation. This approach is less complicated on the decoder side, as compared to previous approaches, since the distribution of pixels need not be determined at the decoder. Moreover, this approach may be less complex than known techniques in HEVC, because fewer classes are used in some groups.

The algorithm shown in Fig. 18 performs a full rate-distortion-based selection of all band offset parameters-the size S of the class, the range R, and the location of the value representing the center C. [ In order to limit the complexity, it is possible to fix certain parameters. In one particular implementation of the algorithm of FIG. 18, the size S and the range R are fixed to the given values known to the encoder and decoder. For example, S may represent 8 pixel values, and R may represent 32 pixel values corresponding to 4 classes of 8 pixels. As a result, the only parameter to be optimized is a value that represents the center C.

Since embodiments of the present invention are considering re-segmenting pixel values over a range of pixel values to determine the classification of pixels, the classification may be adapted accordingly to different pixel value distributions. In particular, the classification may be adapted according to the component type of the pixels. For example, in the case of a chroma component pixel set, the pixel values tend to be lower compared to the pixel values of the luma chroma pixels. In addition, the chroma U pixel values are more concentrated and have a distribution that is different from the distribution of chroma V pixel values with relatively higher pixel values. Moreover, in the case of chroma component pixels, the distribution of pixel values tends to be more concentrated near the peak pixel values, as compared to the distribution of luma chroma pixels providing a more widely diffused distribution.

As can be seen, in addition to the SAO type (no SAO, edge offset, or band offset) and compensation offset values, the parameters S, R, and C, in addition to the SAO type Stream. When the class size and range are fixed, only C is sent to allow the decoder to search the center of the range.

In the case of fixed S and R, one known solution for encoding SAO parameters is to apply the pseudocode of Fig. 20A, which is described in the form of the flowchart of Fig.

This process includes storing the type of SAO (stored in the codeword sao_type_idx), a value indicating the center of the effective range when the band offset type is used (stored in the codeword sao_band_position), and SAO offsets (stored in the codewords sao_offset) Lt; RTI ID = 0.0 > SAO < / RTI > parameters. 20A, cldx represents the index of the color component to which the SAO is applied, rx and ry represent the position of the region to which the SAO is applied, and i is the index of the class of sample values.

The SAO parameter encoding then begins by encoding the SAO type using the unsigned Exp Golomb code (ue (v)) (i.e., unsigned variable length code) in step 2003. If the SAO type is type 5 (band offset), then in step 2017 the encoding continues encoding the value indicating the position in the middle of the validity range using an unsigned fixed length code u (5) of size 5 . The encoding of the four offsets corresponding to the four classes included in the range is then repeated in steps 2019 through 2025. [ Here, each offset is encoded using a signed Exp Golomb code (se (v)) (i.e., a signed variable-length-coding (VLC) code). The encoding process then ends with step 2027. [

If the SAO type is not a band offset, first check whether the SAO type is SAO-free (no SAO means no offset applied to the peer samples). If no SAO is selected, then in step 2027, the encoding process ends.

Otherwise, continue the iterative encoding of the four edge offsets in steps 2007-2013. Again, at step 2027, the encoding process ends.

VLC code is generally used when the range of values to be expressed is relatively large, but some values in this range are more likely than others. The most likely values are given a short code, while the least likely values are given a long code. The main disadvantage of these codes is that they result in higher decoding complexity than FLC (fixed length code). In practice, the VLC code must be read bit-by-bit because it does not know the final size of the code, whereas the FLC code can be read directly because it knows its size.

In Fig. 20B and Fig. 22, an alternative to this encoding process, which replaces the VLC code with the FLC code, is proposed.

The encoding process begins with step 2201, which is the same as step 2001. In step 2203, the VLC coding of the codeword sao_type_idx is replaced with FLC encoding. Three bits are needed here to encode the six possible SAO type values (i.e., "no SAO" type, four "edge offset" types, and "band offset" types). Then, it is checked whether the type of SAO is "no SAO ". In that case, nothing is encoded anymore, and the process ends with step 2215. Otherwise, it is checked whether the type of SAO is "band offset ". In case of YES, a value indicating the position of the center of the range is encoded into a code word SAO_band_position in the form of an unsigned FLC code of size 5. Indeed, in this example, if the class size is 8 sample values and the range consists of 4 classes, 28 different positions are possible for the entire range of 256 values.

Following this step, there will be encodings of four SAO offsets in steps 2211 through 2213. Here, the FLC code replaces the VLC code in steps 2023 and 2011. [ Instead of using up to 5 bits of VLC code covering integer offset values from -31 to 32, here only four different values (generally, (-2, -1, 1, 2) It uses 2 bit size FLC code. Reducing the number of possible values has the effect of focusing the encoding on the most frequently used offset values.

This process ends after offset encoding in step 2215. [

Note that in other embodiments, the range represented by the offsets may be extended by a multiplication factor to be applied to the offsets obtained by the encoding in the picture header, slice header or LCU header, 2-bit code. For example, if the multiplication factor is 4, the encoded offsets (-2, -1, 1, 2) are (-8, -4, 4, 8). The multiplication factor may also be normalized (fixed) or inferred from other LCUs. For example, it can be assumed that the multiplication factor applicable to the previous LCU is applied to the current LCU.

Similarly, in other embodiments, a transition value encoded in a picture header, slice header, or LCU header may be applied to offsets obtained by a two-bit code. For example, if the transition value is 5, the encoded offsets (-2, -1, 1, 2) are (3, 4, 6, 7). Again, the transition value may also be normalized (fixed) or inferred from other LCUs. For example, it can be assumed that a transition value applicable to the previous LCU is applied to the current LCU.

From testing it has been found that having fewer possible offset values does not significantly reduce the performance of the SAO method. It appears that the loss caused by the suppression of certain offset values is compensated by the suppression of the excessive bitrate cost of the less likely offset values.

From additional testing it has also been found that the number of different offset values can be further reduced to three and even two offsets (only one bit is encoded), with very little performance loss.

In Figure 23, further improvements in the encoding process are proposed. Here, it is assumed that the offset values used in the case of the edge offset type can be inferred directly from the type of the edge offset. In that case, the encoding of the edge offset values is not necessary. In other words, each type of edge offset is associated with four classes according to the signal direction, and each class has an associated offset value. This embodiment is triggered by the test showing that, for a given edge offset type and a given class, the offset values are generally close to each other and generally the same. As a result, it is proposed to fix a set of four offset values for each edge offset type. For example, we suggest the following association:

수직 에지 오프셋: (-2, -1, 1, 2)

Vertical edge offset: (-2, -1, 1, 2)

수평 에지 오프셋 (-2, -1, 1, 3)

Horizontal edge offset (-2, -1, 1, 3)

제1 대각 에지 오프셋 (-3, -2, -1, 1)

The first diagonal edge offset (-3, -2, -1, 1)

제2 대각 에지 오프셋 (-1, 1, 2, 3)

The second diagonal edge offset (-1, 1, 2, 3)

The steps 2301 and 2303 in FIG. 23 are the same as the steps 2201 and 2203 already described with reference to FIG. In steps 2305 and 2307, it is checked whether the SAO type is "edge offset" or "no SAO ", respectively. In both of these cases, no offset is encoded. When the SAO type is "edge offset ", upon reading the edge offset type value, the decoder will deduce the offset value from the edge offset type by virtue of the known association with the fixed offset value.

In the embodiment of FIG. 23, if the SAO type is "band offset ", then in step 2309, a value indicating the position in the middle of the range is encoded and the four offset values are repeated by steps 2311 through 2317 / RTI > The encoding process ends at step 2319. [

In the embodiment of Figures 24 and 25, other modifications of the encoding process described in Figures 20A and 21 apply. As already mentioned in the previous embodiment, a 5 bit FLC code is used to encode information (sao_band_position) indicating the position of the center of the range, whereas only 28 different positions are used. In that condition, the four FLC codes (each 5 bits long) remain unused. Here we propose to use these four extra FLC codes to remove the codewords used to encode the SAO type. A new codeword SAO_band_position_and EO will be used to jointly code the range positions and edge offset types. The new codeword is also 5 bits long.

As before, this process begins with defining the SAO parameters in step 2501. This process then encodes the 1-bit length flag (SAO_LCU_flag) indicating whether or not SAO is being used in step 2503. If SAO is not used (step 2505), the process ends (step 2507).

If SAO is used, then in step 2509 it is checked what type of SAO is used. If the SAO type is "band offset ", then in step 2513, the first 28 codes of the codeword SAO_band_position_and_EO are used to encode a value indicating the position in the middle of the range. If the SAO type is "edge offset ", in step 2511, the last four codes of codeword SAO_band_position_and_EO are used to encode the type of edge offset (vertical, horizontal, first diagonal or second diagonal). Following steps 2511 or 2513, four offset values are encoded by steps 2515 to 2521. [

It will be appreciated, however, that in this embodiment redundant codewords are used to encode the type of edge offset, but extra codewords may be used for other purposes, as an alternative. Any other information that needs to be transmitted from the encoder to the decoder may be encoded using the extra codewords.

In Fig. 26, a further embodiment for determining the offsets to be applied in the case of band offsets is proposed. This embodiment is triggered by the test showing that in the case of a band offset, most of the offsets have a low amplitude in absolute value. In practice, the offset values are typically -2, -1, 1, and 2. For example, as in Figures 20A-25, when the number of classes in the range is reduced to four, the number of different groups of four offset values is also reduced. In the example with four different offset values and four classes, the number of different groups is 4 ⁴ = 256. In the embodiments of Figures 21, 22, 23 and 25, eight bits are used to encode offset values (four offset values, each encoded using two bits). Here, all the groups of four offsets are considered to have the same probability of being selected. However, some of these groups are less likely than others. By eliminating less likely groups, it is possible to reduce the number of bits needed to encode them. As a result, instead of encoding four different offset values using two bits for each offset value, assigning indexes to different groups of four offset values and encoding the index - And that the index is encoded using less than 8 bits. The probability of groups can be determined by applying SAO to the training sequence set and calculating statistics for the groups. A table that collects all possible groups that are sorted according to the probability of being selected may be predetermined and known to the encoder and decoder. In this table, each offset group will be associated with an index value. The number of bits assigned to the group index encoding may be fixed (normalized) or fixed for a sequence, frame, slice or LCU, and encoded in corresponding headers. A subset of the groups in the table corresponding to the most probable groups will be determined by the encoder and decoder depending on the number of bits assigned to the index encoding.

An embodiment showing the selection of the best group is described in Fig. This process begins at step 2601 where a variable j is initialized. This variable j is incrementally increased to enable testing of all possible offset groups. In the proposed embodiment, a group of four offsets is contemplated, but other offsets may be conceivable. In step 2603, it is checked whether all the groups have been tested (for example, NbOffsetGroup can be 128). If yes, the process ends and less than 8 bits of codeword corresponding to the selected group are encoded. If not, the process continues with the initialization of variable i, allowing all classes in the range to be tested (step 2605). In step 2606, a variable SumDiff (j) representing the sum of the differences between the original samples and the SAO filtered encoded samples corresponding to the offset group j is initialized to zero. Here, only four classes are considered, but the number of different classes matching the number of offsets is possible. In step 2607, if there remain some classes to be tested, the variable k is initialized to allow testing of all possible samples in the sample range corresponding to class i. The steps 2611 to 2619 calculate the sum of the absolute values of the differences between the encoded samples filtered at offset of class i in offset group j and the original samples in class i considered. Where orig (k) is the average of the original sample values corresponding to the encoded value enc (k). filter (i, j) is an offset value corresponding to class i in offset group j. SumDiff (j) is the sum of the differences computed for all classes (here, four classes) that make up the range. In the loop comprising steps 2621 through 2626, the sums of all calculated differences are compared and the index of the group with the smallest sum is selected. The selected index corresponds to an offset group that, when applied to the encoded samples, allows to minimize the difference between the filtered samples and the original sample.

During the syntax encoding process, for example, the encoding of the offset values indicated by steps 2211 through 2213 of Figure 22 is replaced with the encoding of the index corresponding to the selected offset group.

The classification of the luma component pixels is performed separately from the classification of the chroma U or V component pixels and thus each classification can be adapted accordingly so that each class has a similar number of pixels.

Thus, the methods of embodiments of the present invention provide a more flexible classification scheme that can be configured to provide a more optimal classification independently for both luma or chroma signals, thereby resulting in improved coding efficiency.

Although the present invention has been described above with reference to specific embodiments, it is to be understood that the invention is not limited to the specific embodiments, and modifications that are within the scope of the invention will be apparent to those skilled in the art.

For example, while previous embodiments have been described with reference to pixels of an image and their corresponding pixel values, it should be appreciated that, in the context of the present invention, a group of pixels may be considered with corresponding group pixel values . The sample may thus correspond to one or more pixels of the image.

Additional aspects of the invention are described below.

According to a first additional aspect of the present invention there is provided a method of providing compensation offsets for a reconstructed sample set of images, each sample having its own sample value, The method comprising the steps of determining a plurality of classes for re-segmenting samples according to a corresponding sample value of samples, depending on the characteristics of the statistical distribution of sample values, Define a range of sample values of; And associating each of the determined classes with a respective compensation offset for applying to a respective sample value of each of the samples of the class.

In this aspect of the invention, since the statistical distribution of the sample values is taken into account in determining the classification of the samples, the classification may accordingly be adapted to the ranges of all possible sample values. Moreover, the classification can be adapted according to the component type of the samples. For example, in the case of pixels corresponding to a chroma signal, the distribution of pixel values tends to be more concentrated in the vicinity of peak pixel values, as compared to a luma signal providing a more widely diffused distribution. Thus, the methods of embodiments of the present invention provide a more flexible classification scheme that can be configured to provide a more optimal classification independently for both luma or chroma signals, thereby resulting in improved coding efficiency.

In one embodiment, the properties of the statistical distribution include a determined center of statistical distribution of image sample values.

In one embodiment, the characteristics of the statistical distribution include a range of valid sample values of the statistical distribution.

In one embodiment, the classes are determined such that the samples are substantially uniformly shared among the classes.

In one embodiment, the samples of the set may be at least a first component type or a second component type, and a plurality of classes are determined depending on the component type of the sample set.

In one embodiment, the number of classes is predetermined and the range of sample values defined for each class is determined depending on the properties of the statistical distribution.

In another embodiment, the number of classes and the range of sample values defined for each class are determined depending on the properties of the statistical distribution.

In one embodiment, the number of classes is determined depending on the number of samples having sample values within the valid range.

In one embodiment, the number of classes is determined according to the component type.

In one embodiment, the method includes determining a maximum number of sample values per class in accordance with the number of sample values in the validity range.

In one embodiment, the step of determining a plurality of classes comprises selecting, from among a plurality of predetermined classifications, a class defining a plurality of classes that are adapted to the properties of the statistical distribution.

In one embodiment, the range of sample values for each class, the center of the statistical distribution, and / or the range of valid sample values are determined based on the rate distortion criterion.

In one embodiment, the range of sample values for each class, the center of the statistical distribution, and / or the range of valid sample values is predetermined. In one embodiment, if the range of valid sample values is determined based on a comparison of sample values for a threshold, the threshold depends on the total number of samples, the threshold depends on the component type of the samples, Is the determined value.

In one embodiment, the compensation offset for each class is determined from an average of the difference between the sample value of each reconstructed sample of the class and the corresponding sample value of the original image.

In one embodiment, the sample set is one of a plurality of sample sets of images, and the same number of classes is determined for each set.

In one embodiment, the sample value represents the bit depth, and the effective range, the range of each class, and / or the center of the statistical distribution is dependent on the bit depth.

In one embodiment, the range of sample values for a given class depends on the location of the class within its validity range.

In one embodiment, the range of sample values for a given class located at the edge of the statistical distribution is greater than the range of sample values for a given class in the central region of the distribution.

In one embodiment, the center of the statistical distribution is determined based on the effective range.

In one embodiment, the plurality of classes comprises a first class group located at a central portion of the statistical distribution and a second class group comprising first and second subclass groups located at respective edge portions of the statistical distribution.

In one embodiment, the positions of the second subclass groups in the statistical distribution are provided as data for coding.

In one embodiment, the positions of the subgroups of the second group are independent of the full extent of the statistical distribution.

In one embodiment, the center of the statistical distribution is not provided for coding.

In one embodiment, the position of the effective range (classification) within the entire range is selected from a plurality of possible positions distributed over the entire range, and the interval between two consecutive positions is at least one Section is smaller than in the other parts of the entire range.

For example, a portion with a smaller spacing may be in the middle of the entire range.

According to another example, a portion having a smaller spacing may be at one or both ends of the full range.

The positions may be centered. Alternatively, the locations may be end locations.

Possible locations can be assigned indices.

In one embodiment, the range of sample values for each class, the center of the statistical distribution, and / or the range of valid sample values is predetermined.

In one embodiment, at least two different types of offset values may be generated and one of the types is so-called band offset values, each of which is associated with classes (ranges of sample values), as described above . One or more other types of offset values may be so-called edge offset values. There can be different types of edge offset values for each direction - for example, four different types correspond to four different directions, respectively. As such, there may be a set of edge offset values for each direction. One type of offset value (band or edge) can then be selected. The selection criterion is not limited, and one suitable criterion is the rate distortion criterion. Only the selected type of SAO filtering (band or edge offset) is applied by the encoder and decoder. Selection may be performed for each frame region. In addition, it may be possible to select "no SAO ", i. E. To apply neither band offset nor edge offset, if much improvement is not achieved, for example by either type. In one embodiment, the selected type of SAO filtering is encoded by the encoder and sent to the decoder.

In one embodiment, a fixed length code is used to encode the selected type.

In one embodiment, when the selected type is a band offset, a value indicating the location of the validity range is encoded. The position can be the center position or the end position of the effective range.

In one embodiment, a fixed length code is used to encode the location of the validity range.

In one embodiment, when the number of different code words allowed by the length of the fixed length code used to encode the codewords corresponding to the values of the positions of the valid range is greater than the actual number of different possible positions, Extra codewords are used to encode. For example, in one embodiment, extra code words are used to encode the type (direction) of edge offsets.

In one embodiment, compensation offsets are encoded with fixed length codes. This can be applied to both band offsets or edge offsets or offsets of these types.

In one embodiment, the number of different compensation offsets allowed by the length of the fixed length code is less than the number of possible compensation offsets.

In one embodiment, a multiplication factor is applied to the compensation offset values, and the compensated offset values used for filtering are used.

In one embodiment, the multiplication factor is sent from the encoder to the decoder.

In one embodiment, a shifting value is applied to the compensation offset values, and the compensated offset values used for filtering are used.

In one embodiment, the transition value is sent from the encoder to the decoder.

In one embodiment, edge offset values for at least one type (direction) are predetermined.

In one embodiment, a flag is encoded that indicates whether compensation offsets need to be applied to the reconstructed sample set of images.

In one embodiment, the band offset value groups are predefined, and each offset of the group is associated with one class of validity range.

In one embodiment, each group is assigned an index.

In one embodiment, the predefined groups are collected in a table, and the table order depends on the selection probability of each group.

In one embodiment, the selection probabilities of each group are calculated by applying methods to provide compensation offsets implementing the present invention to a training sequence set.

In one embodiment, a group of band offset values among predefined band offset value groups having a minimum difference between the original samples and the filtered samples is selected.

In one embodiment, the indices of the groups are encoded using a fixed length code.

In one embodiment, the number of different encoded group indices allowed by the length of the fixed length code is less than the number of different possible groups.

In one embodiment, a subset of the band offset groups within the different possible groups is determined depending on the selection probability of each group and the length of the fixed length code.

In one embodiment, the length of the fixed length code used to encode the indices of the band offset groups is predetermined.

In one embodiment, the length of the fixed length code used to encode indices of compensating offset groups is encoded into a sequence header, a picture group header, a picture header, a slice header, or an LCU header.

According to a second additional aspect of the present invention, there is provided a method of encoding an image comprising a plurality of samples, the method comprising: encoding samples; Decoding the encoded samples to provide reconstructed samples; Performing loop filtering on reconstructed samples, wherein the loop filtering comprises applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Are provided according to a method of a first additional aspect; And generating a bit stream of encoded samples.

In one embodiment, the method includes transmitting, in the bitstream, encoded data representing respective compensation offsets for each class.

In one embodiment, the method includes transmitting, in a bitstream, encoding classification data defining a plurality of determined classes.

In one embodiment, the classification data includes data indicative of the center of the statistical distribution.

In one embodiment, the classification data includes data representing a range of sample values defined for each of a plurality of classes.

In one embodiment, the classification data includes data indicative of an effective range of the statistical distribution.

According to a third further aspect of the present invention there is provided a method of decoding an image consisting of a plurality of samples,

Receiving encoded samples;

Decoding the encoded samples to provide reconstructed samples; And

Performing loop filtering on reconstructed samples, wherein the loop filtering comprises applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Are provided according to any of the previous embodiments.

Another aspect of additional aspects of the present invention provides a method of decoding an image comprising a plurality of sample values,

Receiving the encoded sample values;

Receiving encoded classification data defining a plurality of classes associated with respective compensation offsets provided according to the method of any of the preceding embodiments, wherein each compensation offset corresponds to a range of sample values -;

Decoding the encoded samples and decoding the encoded compensation offsets to provide reconstructed samples; And

Performing loop filtering on the reconstructed samples according to a sample value of the sample, wherein the loop filtering comprises applying the received compensation offsets to the image samples of the respective samples.

In one embodiment, the classification data includes data representing the center of the statistical distribution, data representing a range of sample values defined for each of the plurality of classes, and / or data representing the valid range of the statistical distribution.

According to another of its further aspects, there is provided a signal carrying an information data set for an image represented by a video bitstream, the image comprising a set of reconstructable samples, each reconstructable sample comprising After reconstruction, has its own sample value, the sample values of the plurality of reconstructed samples of the set being representable as a statistical distribution of sample values; The information data set includes classification data representing a plurality of classes associated with respective compensation offsets for applying to sample values of respective reconstructed samples, and the classification data is determined according to a statistical distribution.

In one embodiment, the classification data includes data representing the center of the statistical distribution, data representing a range of sample values defined for each of the plurality of classes, and / or data representing the valid range of the statistical distribution.

According to another of its further aspects, there is provided an apparatus for providing compensation offsets for a reconstructed sample set of images, each sample having its own sample value, Can be represented as a statistical distribution of sample values,

Means for determining a plurality of classes for subdividing samples according to a corresponding sample value of samples, depending on characteristics of statistical distribution of sample values, each class defining a range of respective sample values; And

And means for associating each of the determined classes with a respective compensation offset for applying to a respective sample value of each sample of the class.

Another aspect of additional aspects of the present invention provides an encoding apparatus for encoding an image comprising a plurality of samples,

An encoder for encoding samples; A decoder for decoding the encoded samples to provide reconstructed samples; Loop filter-loop filtering means for filtering reconstructed samples comprises offset applying means for applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Are provided by the apparatus of the previous embodiment; And a bitstream generator for generating a bitstream of encoded samples.

Yet another aspect of additional aspects of the present invention provides a decoding apparatus for decoding an image comprising a plurality of samples, the apparatus comprising: a receiver for receiving encoded samples; A decoder for decoding the encoded samples to provide reconstructed samples; And a loop filter-loop filter loop-filtering the reconstructed samples comprises means for applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Provided by an apparatus according to the previous embodiment.

Another aspect of additional aspects of the present invention provides a decoding apparatus for decoding an image consisting of a plurality of sample values, the apparatus comprising: means for receiving encoded sample values and providing A receiver for receiving encoded classification data defining a plurality of classes associated with respective compensation offsets, each compensation offset corresponding to a range of sample values; A decoder to decode the encoded samples and to decode the encoded compensation offsets to provide reconstructed samples; And a loop filter-loop filter that loops the reconstructed samples according to a sample value of the sample, comprising means for applying received compensation offsets to image samples of respective samples.

Many additional modifications and variations will be apparent to those skilled in the art with reference to the above description of exemplary embodiments, which are provided by way of example only and are not intended to limit the scope of the invention, which is determined solely by the appended claims. In particular, the different features from the different embodiments can be interchanged, if appropriate.

In the claims, the term " comprising "does not exclude other elements or steps, and the singular expression" one " The mere fact that different features are recited in different dependent claims does not mean that a combination of these features can not be used to advantage.

Claims (47)

A method of providing compensation offsets for a set of reconstructed samples of an image,
Each sample having a sample value, the method comprising:
Selecting a classification from among a plurality of predetermined classifications based on a rate distortion criterion, each of the predetermined classifications having a classification range that is less than the entire range of sample values, Wherein each class defines a range of sample values within the classification range and the sample is placed in the class if the sample value of the sample is within the range of the associated class; And
And associating a compensation offset for applying to a sample value of each sample of the associated class with each class of the selected classification. 2. The method of claim 1, wherein the classification range of one or more of the predetermined classifications is less than one-half of the entire range. 2. The method of claim 1, wherein the classification range of one or more of the predetermined classifications is 1/8 of the entire range. 2. The method of claim 1, wherein one or more of the predetermined classifications comprises four classes of compensating offsets for a set of reconstructed samples of an image. 5. A method according to any one of claims 1 to 4, wherein the samples of the set can be at least a first component type or a second component type and the plurality of classes are determined depending on the component type of the sample set And compensating offsets for a set of reconstructed samples of the image. 6. The method of claim 5, wherein the number of classes is determined according to the component type. 7. The method of any one of claims 1-6, wherein the compensation offset for each class is calculated by multiplying the average of the differences between the sample value of each reconstructed sample of the class and the respective sample value of the corresponding original image ≪ / RTI > of the reconstructed samples of the image. 8. The method according to any one of claims 1 to 7, wherein the sample value represents bit depth and wherein at least one of the classification range, the range of each class and the center of the selected classification is dependent on the bit depth. A method for providing compensation offsets for a set of reconstructed samples of an image. 9. A method as claimed in any one of claims 1 to 8, wherein the range of sample values for a given class is dependent on the location of the class in the selected class. How to provide. 10. The method of claim 9, wherein a range of sample values for a given class located at an edge of the selected class is greater than a range of sample values for a given class in a central region of the selected class. Lt; / RTI > CLAIMS 1. A method of encoding an image comprising a plurality of samples,
Encoding the samples;
Decoding the encoded samples to provide reconstructed samples;
Performing loop filtering on the reconstructed samples, the loop filtering comprising applying compensation offsets to sample values of respective reconstructed samples, each compensation offset being associated with a range of sample values, Wherein the compensation offsets are provided according to the method of any one of claims 1 to 10; And
And generating a bit stream of encoded samples. 12. The method of claim 11, further comprising transmitting, in the bitstream, encoded data representing the compensation offsets that are each associated with a plurality of classes of a selected classification. 14. The method of claim 11 or 12, further comprising transmitting, in the bitstream, encoded classification data about the selected classification. 14. The method of claim 13, wherein the classification data comprises data representing a center of the selected classification. 14. The method of claim 13, wherein the classification data comprises data indicative of an index associated with the selected classification. 14. The method of claim 13, wherein the classification data comprises data representing a location of the selected classification within a full range of sample values. 17. The method of claim 16, wherein the location is an end location of the selected taxon. 18. The method of claim 17, wherein the end position is represented as a displacement from one end of the entire range. 19. The method according to any one of claims 13 to 18, wherein the classification data comprises data representing a range of sample values defined for each of the plurality of classes. 20. The method according to any one of claims 13 to 19, wherein the classification data comprises data representing a classification range of the selected classification. CLAIMS 1. A method for decoding an image comprising a plurality of samples,
Each sample having a sample value, the method comprising:
Receiving the encoded sample values;
Receiving encoded classification data;
Receiving encoded compensated offsets;
Decoding the classification data and selecting a classification from among a plurality of predetermined classifications based on the decoded classification data, each of the predetermined classifications having a classification range smaller than the entire range of sample values, Each class defining a range of sample values within the classification range, and if the sample value of the sample is within the range of the associated class, the sample is placed in the class;
Decoding the encoded samples to provide reconstructed sample values, and decoding the encoded compensation offsets;
Associating the decoded compensation offsets with each of the classes of the selected classification; And
Performing loop filtering on the reconstructed sample values, wherein the loop filtering comprises applying the decoded compensated offset associated with each class of the selected class to reconstructed sample values within the range of the class of interest - < / RTI > 22. The method of claim 21, wherein the classification range of one or more of the predetermined classifications is less than one-half of the entire range. 22. The method of claim 21, wherein the classification range of one or more of the predetermined classifications is 1/8 of the entire range. 23. The method of claim 21, wherein one or more of the predetermined classifications comprises four classes. 25. An image decoding method according to any one of claims 21 to 24, wherein the classification data includes data indicating a center of the selected classification. 25. An image decoding method according to any one of claims 21 to 24, wherein the classification data includes data representing an index related to the selected classification. 25. An image decoding method according to any one of claims 21 to 24, wherein the classification data includes data representing the position of the selected classification within the entire range of sample values. 28. The method of claim 27, wherein the position is an end position of the selected classification. 29. The method of claim 28, wherein the end position is represented as a displacement from one end of the entire range. 30. An image decoding method according to any one of claims 21 to 29, wherein the classification data includes data indicating a range of sample values defined for each of the plurality of classes. 31. The method according to any one of claims 21 to 30, wherein the classification data includes data representing a classification range of the selected classification. 31. A computer program product for a programmable apparatus, comprising: a computer program that when executed by the programmable apparatus is loaded into the programmable apparatus and comprises a sequence of instructions implementing the method of any one of claims 1 to 31 product. 33. A computer-readable storage medium storing instructions of a computer program for implementing the method of any one of claims 1 to 32. A signal carrying an information dataset for an image represented by a video bitstream,
Wherein the image comprises a set of reconstructible samples, each reconstructable sample having a sample value, the information data set comprising classification data relating to a classification selected by the encoder from among a plurality of predetermined classifications, The predetermined classifier has a classification range smaller than the entire range of sample values and is made up of a plurality of classes, each class defines a range of sample values within the classification range, and the sample value of the sample is within the range of the related class The sample is placed in the class and each class of the plurality of classes of the selected class is associated with a compensation offset for applying to sample values of the reconfigurable samples in the range of the class of interest. 35. The signal of claim 34, wherein the classification range of one or more of the predetermined classifications is less than one half of the entire range. 35. The signal of claim 34, wherein the classification range of one or more of the predetermined classifications is one-eighth of the full range. 35. The signal of claim 34, wherein the one or more of the predetermined classifications consists of four classes. 37. The signal according to any one of claims 34 to 37, wherein the classification data comprises data indicative of the center of the selected classification. 37. The signal according to any one of claims 34 to 37, wherein the classification data comprises data indicative of an index associated with the selected classification. 37. The signal according to any one of claims 34 to 37, wherein the classification data comprises data representing the location of the selected classification within the entire range of the sample value. 41. The signal according to claim 40, wherein the position is an end position of the selected classification. 42. The signal according to claim 41, wherein the end position is represented as a displacement from one end of the entire range. 43. The signal according to any one of claims 34 to 42, wherein the classification data includes data representing a range of sample values defined for each of the plurality of classes. 44. The signal according to any one of claims 34 to 43, wherein the classification data comprises data representing an effective range of the selected classification. An apparatus for providing compensation offsets for a set of reconstructed samples of an image,
Each sample having a sample value, the apparatus comprising:
Means for selecting a classification from among a plurality of predetermined classifications based on a rate distortion criterion, each said predetermined class having a classification range smaller than the entire range of sample values and comprising a plurality of classes, Define a range of sample values within the classification range, and if the sample value of the sample is within the range of the associated class, the sample is placed within the class; And
And means for associating a compensation offset for applying to a sample value of each sample of an associated class with each class of the selected classification. An encoding apparatus for encoding an image comprising a plurality of samples,
An encoder to encode the samples;
A decoder to decode the encoded samples to provide reconstructed samples; And
A loop filter for filtering the reconstructed samples, the loop filtering means comprising offset applying means for applying compensation offsets to sample values of the respective reconstructed samples, each compensation offset being associated with a range of sample values The compensation offsets provided by the apparatus of claim 45; And
And a bitstream generator for generating a bitstream of encoded samples. An apparatus for decoding an image comprising a plurality of samples,
Each sample having a sample value, the apparatus comprising:
Means for receiving encoded sample values;
Means for receiving encoded classification data;
Means for receiving encoded compensated offsets;
Means for decoding the classification data and selecting a classification from among a plurality of predetermined classifications based on the decoded classification data, each of the predetermined classifications having a classification range smaller than the entire range of sample values, Wherein each class defines a range of sample values within the classification range and the sample is placed in the class if the sample value of the sample is within the range of the associated class;
Means for decoding the encoded samples to provide reconstructed sample values, and decoding the encoded compensation offsets;
Means for associating the decoded compensation offsets with each of the classes of the selected classification; And
Means for performing loop filtering on the reconstructed sample values, wherein the loop filtering comprises applying the decoded compensation offset associated with each class of the selected class to reconstructed sample values in the range of the class of interest. - < / RTI >

Images